DanielMalvezziDoine ForeignExchangeInterventionsandCovered ...€¦ · DanielMalvezziDoine...

Daniel Malvezzi Doine

Foreign Exchange Interventions and CoveredInterest Parity Deviations

Dissertação de Mestrado

Dissertation presented to the Programa de Pós–graduação emEconomia da PUC-Rio in partial fulfillment of the requirementsfor the degree of Mestre em Economia .

Advisor: Prof. Márcio Gomes Pinto Garcia

Rio de JaneiroAugust 2020

DBD

PUC-Rio - Certificação Digital Nº 1811825/CA


Foreign Exchange Interventions and CoveredInterest Parity Deviations

Dissertation presented to the Programa de Pós–graduação emEconomia da PUC-Rio in partial fulfillment of the requirementsfor the degree of Mestre em Economia . Approved by theExamination Committee.

Prof. Márcio Gomes Pinto GarciaAdvisor

Departamento de Economia – PUC-Rio

Prof. Marcelo Cunha MedeirosDepartamento de Economia – PUC-Rio

Marcos de Carvalho ChamonResearch Department – International Monetary Fund

Rio de Janeiro, August the 24th, 2020

DBD


All rights reserved.


B.A., Economics, Universidade de São Paulo, 2016

Bibliographic dataDoine, Daniel Malvezzi

Foreign Exchange Interventions and Covered Interest Pa-rity Deviations / Daniel Malvezzi Doine; advisor: Márcio Go-mes Pinto Garcia. – Rio de janeiro: PUC-Rio , Departamentode Economia, 2020.

v., 55 f: il. color. ; 30 cm

Dissertação (mestrado) - Pontifícia Universidade Católicado Rio de Janeiro, Departamento de Economia.

Inclui bibliografia

1. Economia – Teses. 2. Economia – Teses. 3. Intervençõescambiais;. 4. Paridade coberta da taxa de juros;. 5. ArtificialCounterfactual;. 6. Local Projection;. 7. Taxa de câmbio;.8. Taxa de câmbio futura;. 9. Derivativos;. I. Garcia, MárcioGomes Pinto. II. Pontifícia Universidade Católica do Rio deJaneiro. Departamento de Economia. III. Título.

CDD: 620.11

DBD


Acknowledgments

Several people are involved in the process of elaborating on this thesis, andeach one has a unique contribution.

First of all, I would like to thanks my advisor Prof. Márcio Garcia, forthe excellent guidance, friendship, and advice, which were very important inmy formation.

I also thank Prof. Marcelo Medeiros and Marcos Chamon for importantcomments, suggestions, and insights.

During my time at PUC, I appreciated the classes of the professors Mar-celo Medeiros, Fábio Miessi, Ruy Ribeiro, Eduardo Zilberman, and Marcelode Paiva Abreu. It was a privilege to spend some time listening and absorbingpart of your knowledge.

This project would not be possible without the support of my family, myparents Mauro and Rita, my sister Sofia, and my grandparents, Mauro andMaria.

Finally, I am very grateful to all my friends and my classmates. My periodin Rio would not be the same without your friendship.

This study was partially financed by the Coordenação de Aperfeiçoa-mento Pessoal de Nível Superior – Brasil (CAPES) – Código de Financiamento001. I also thanks the CNPq for the financial support.

DBD


Abstract

Doine, Daniel Malvezzi; Garcia, Márcio Gomes Pinto (Advisor). Fo-reign Exchange Interventions and Covered Interest ParityDeviations. Rio de Janeiro, 2020. 55p. Dissertação de mestrado– Departamento de Economia, Pontifícia Universidade Católica doRio de Janeiro.

Traditionally, much has been written about the effects of FX (foreignexchange) sterilized interventions on exchange rates, both theoretically andempirically, with mixed results. More recently, the international financeliterature has tried to explain the deviations from the well-known CoveredInterest Parity (CIP) condition that have, since the 2008 Great FinancialCrisis, arisen among advanced economies currencies. Here, we originallymerge these two strands of the literature by analyzing the effects of sterilizedFX interventions on the CIP (Covered Interest Parity) deviation. Oursample is composed of Brazilian Central Bank FX interventions between2009 and 2020. This period contains a major program of announced FXinterventions in response to the Taper Tantrum, in 2013, which has alreadybeen shown to have significantly affected the level of the exchange rate(Chamon, Garcia, and Souza (2017)). To gauge the effects, we build acounterfactual employing the ArCo methodology, developed by Carvalho,Masini, and Medeiros (2018), and also make use of Jordà (2005) LocalProjections. The results indicate that selling US dollars in the futuresmarket increases CIP deviations while buying US dollar futures has theopposite effect. Offering US dollar repo credit lines points to a short-liveddecrease in the deviation. The number of sterilized sales or purchases ofspot currency seems not to be high enough to lead to conclusive results.

KeywordsForeign exchange intervention; Covered interest parity; Artificial

Counterfactual; Local Projection; Exchange rate; Forward exchangerate; Derivatives;

DBD


Resumo

Doine, Daniel Malvezzi; Garcia, Márcio Gomes Pinto. Interven-ções Cambiais e Desvios na Paridade Coberta da Taxa deJuros. Rio de Janeiro, 2020. 55p. Dissertação de Mestrado – De-partamento de Economia, Pontifícia Universidade Católica do Riode Janeiro.

Tradicionalmente, muitos trabalhos têm estudado os efeitos das inter-venções cambiais esterilizadas nas taxas de câmbio, tanto empiricamentequanto teoricamente, encontrando resultados mistos. Mais recentemente, aliteratura de finanças internacionais têm procurado explicar os desvios naParidade Coberta da Taxa de Juros (PCJ), que vem sendo observado en-tre as moedas das economias desenvolvidas após a Grande Crise Financeirade 2008. Neste trabalho, ligamos as duas literaturas ao estudar o efeito dasintervenções cambiais nos desvios na paridade coberta de juros. Nossa amos-tra consiste nas intervenções realizadas pelo Banco Central do Brasil entreos anos de 2009 e 2020. Este período contempla o programa de interven-ções pré-anunciadas de 2013, implementado no contexto do Taper Tantrum,e que já mostrou ter afetado significantemente as taxas de câmbio (Cha-mon, Garcia e Souza (2017)). Para avaliar os efeitos, construímos uma sériecontrafactual utilizando a metodologia ArCo, desenvolvida por Carvalho,Masini e Medeiros (2018), e também estimando funções impulso respostautilizando Local Projection, desenvolvida por Jordà (2005). Os resultadosindicam que a venda de dólares no mercado futuro aumentam os desviosna PCJ, enquanto que compras de dólares tem o efeito oposto. A oferta dedólares via contratos de recompra diminui os desvios no curto prazo. Asintervenções no mercado a vista apresentam resultados inconclusivos.

Palavras-chaveIntervenções cambiais; Paridade coberta da taxa de juros; Artificial

Counterfactual; Local Projection; Taxa de câmbio; Taxa de câmbiofutura; Derivativos;

DBD


Table of contents

1 Introduction 101.0.1 Related Literature 14

2 Methodology 172.0.1 Artificial Counterfactual (ArCo) 172.0.1.1 Identification 192.0.2 Local Projection 20

3 Data 233.0.1 The Program of Daily FX Intervention 233.0.2 CIP Deviations 253.0.2.1 Brazil 263.0.3 Interventions 27

4 Results 304.0.1 Daily ArCo 314.0.2 Weekly ArCo 344.0.3 Local Projections 36

5 Conclusion 41

A Data Appendix 45

B Results Appendix 46

DBD


List of figures

Figure 1.1 CIP Deviation and Swap Interventions 13

Figure 3.1 Periods of Analysis 25Figure 3.2 Foreign Exchange Interventions 28

Figure 4.1 Average Maturity of the New Offered Swap Contracts 32Figure 4.2 Results - Daily ArCo 33Figure 4.3 Results - Weekly ArCo 35Figure 4.4 Swap Interventions Effect on CIP Deviations 36Figure 4.5 Swap Interventions Effect on CIP Deviations - Covid-19 37Figure 4.6 Spot Interventions Effect on CIP Deviations 38Figure 4.7 Reverse Swap Effect on CIP Deviations - 2H2019 39Figure 4.8 Repo Interventions Effect on CIP Deviations 39

Figure B.1 Results - Daily ArCo - Pre 1 48Figure B.2 Results - Daily ArCo - Pre 2 49Figure B.3 Results - Weekly ArCo - Pre 1 50Figure B.4 Results - Weekly ArCo - Pre 2 51Figure B.5 Results - Adjusted R-Squared 52Figure B.6 Swap Interventions Effect on CIP Deviations 53Figure B.7 Swap Interventions Effect on CIP Deviations - Covid-19 53Figure B.8 Spot Interventions Effect on CIP Deviations 54Figure B.9 Reverse Swap Effect on CIP Deviations - 2H2019 54Figure B.10 Repo Interventions Effect on CIP Deviations 55

DBD


List of tables

Table 3.1 CIP Deviations - Descriptive Analyses 27Table 3.2 Samples - Local Projection (LP) 29

Table B.1 List of the Models - Daily Data 46Table B.2 List of the Models - Weekly Data 47

DBD


1Introduction

In the onset of unconventional monetary policies ruled by central banksof developed economies in the period that followed the 2008 Global FinancialCrisis (GFC), the Emerging Market Economies (EME) were the destiny ofa massive flow of capital. This movement leads EME central banks to useforeign exchange interventions (FXI) to accumulate international reserves, inthe period of capital inflow, and to smooth the effects of capital outflow on theexchange rates from 2013 on, in the so-called taper tantrum.

These interventions brought out the question of whereas sterilized FXIcan or cannot affect the exchange rate path. In the last decade, several papershave addressed this question, for different periods, countries, and using severaleconometric techniques, always aiming to circumvent the inherent endogeneityproblem in this literature, related to the fact that central banks intervene asa response to the exchange rate movements. Answering this question is not abrand new desire in international economics. Sarno and Taylor (2001) surveyedthe 1980 and 90’s literature. A survey of the recent FXI ruled by the EME canbe found in Menkhoff (2013). The results vary from country and period butpoint to the effectiveness of those interventions.

A second and more recent wave of papers is investigating a slightlydifferent question of to what extent these FXI produce spillovers over othereconomic variables rather than the exchange rate. Perhaps the first paper tocarry out this question was Blanchard et al. (2015), investigating how capitalflows are affected by the interventions. Hofmann et al. (2019) and Gonzalezet al. (2019), for instance, analyze the effect on credit market.

Beyond the FXI’s, the 2008 GFC also triggered deviations on the CoveredInterest Parity (CIP) among the developed economies’ currencies. In thescenario of lack of risks, any difference between the forward premium andthe interest rate differential, the CIP components, would return a risk-freepositive profit, so any deviation is expected to vanish by players operatingthis wedge. There is flourishing literature investigating the reasons behindthe deviations. Although quite common among emerging markets due to thecountry and convertibility risks, the parity was holding for the G10 currencyhistorically. Among the causes, the most cited one is related to changes in

DBD


Chapter 1. Introduction 11

banking regulation, diminishing bank’s capacity to operate these arbitrageand leading to the deviations. Then, the deviations emerge as a result of a lackof liquidity in the interbank loan market across the countries.

In this paper, we link this two literature and investigate the FXI effectson the CIP deviations. We analyze Brazilian interventions, from 2009 until2020, including the major program of interventions ruled by the BrazilianCentral Bank (BCB) between 2013 and 2015. We employ two different andcomplementary methodologies to capture the CIP dynamics as a response foran intervention: the Artificial Counterfactual (ArCo), developed by Carvalhoet al. (2018), and the Jordà (2005) Local Projection.

To construct the ArCo statistics and evaluate the major swap programeffect on the CIP deviation, we build up a control group composed of countriesthat did not intervene in the FX market between 2013-2015. Besides this verystudied program, we also extend our sample from 2009 until 2020 to includespot and repo interventions. The sample also contemplates recent interventionsruled as a response to the Covid-19 crisis. We then use the Local Projectionsto estimate IRF of CIP deviations responses to interventions shocks. This lastmethodology allows us to measure the exact impact of a specific intervention,as well as study asymmetries between purchasing or selling foreign currency.

The reason we use two methodologies for the problem regards the natureof interventions. In the major program of 2013-2015, the Central Bank ruleddaily pre-announced interventions of an equal amount. Then, any IRF wouldnot capture this constant and continuous shock appropriately, as well as thewhole program effect, as the ArCo does. The drawbacks of ArCo includethe lack of information about a specific intervention and the difficulty inbuilding a control group. On the other hand, the interventions ruled outsidethe program consist of unannounced and different-size operations, the reasonwhy we use Local Projection’s IRF. Differently from the ArCo, which capturesthe average sample effect, LP’s IRF allows us to understand the daily effect ofan intervention. Moreover, expanding the sample period, we can analyze othertypes of interventions used by the BCB: spot and repurchase (repo) credit linesoperations. The main challenge of Local Projection estimations is to defineintervals of estimations since spot and forward interventions are clustered inspecific periods while repo interventions are less clustered but observed in thewhole sample.

The results we found can be divided into three components. First, wefind that forward intervention indeed affects CIP deviations, and not onlyin the short term. A sale of forward currency by a monetary authority putsdownward pressure on the forward prices, shrinking the forward premium and

DBD



widening the parity deviations. The ArCo results indicate that the major swapprogram leads to an increase in the CIP deviations of approximately 35 basispoints, on average. Local Projection impulse response functions’ indicate thata forward intervention leads to an increase of up to 10 basis points on the CIPdeviations, and this effect persists around 20 days. We also find evidence ofsymmetric effect regarding reverse swap operations.

Second, there is evidence that this effect is not significant when lookingat spot market interventions, illustrating a possible difference between thesetwo available instruments on the central bank toolkit. However, we observethis result cautiously since the spot market interventions contemplated in oursample were ruled in specific environments, as explained in Section 3, whichcan weaken the conclusions. The results for repo interventions sales, on theother hand, indicate a decrease in the deviations, especially on the shortermaturity. We believe this pattern is consistent with BCB strategy adopted onthe major swap program, when repo sales were released every week, aiming togive liquidity to the market.

Our findings are robust for several tests. First, using daily data, theArCo is robust for several specifications and lags used. A weakness of the dailyfrequency is the lack of data related to capital flows for most countries in thecontrol group. Then, we add to previous analyses weekly data of bonds andequity flows, and results remain the same for the new data frequency. Lastly,we take advantage of a detailed interventions data, released by the BrazilianCentral Bank, to estimate Local Projection impulse response functions. Theresults for the swap goes in the same line of the ArCo results. Local Projectionalso adds to our study by differentiating long and short intervention’s positionand operations on the spot and repo markets.

To elucidate the variables we are working, Figure 1.1 shows the 12-monthCIP deviation for Brazil and the thirteen countries used as the control groupon the ArCo analysis, as will be detailed in Section 2. The period correspondsto the 2013-2015 major pre-announcement FXI program. The grey area, on theright-hand-side, corresponds to the accumulated offered swap amount. Brazil’sCIP deviation is highlighted and shows some interesting patterns. The mostimportant is the increase right after the program announcement, pointing tothe effect we aim to capture. Second, the Brazilian deviation is higher than itspeers, on average, and increases along with the program.

The mechanism we are studying works as follows. Suppose the BrazilianCentral Bank offer dollars in the forward market aiming to provide liquidity.The increase in the supply puts downward pressure on the forward exchangerates. Once the spot exchange rate and the foreign interest rate remain

DBD



Figure 1.1: CIP Deviation and Swap Interventions

Figure 1.1 shows the 10-day moving average CIP deviations, for the maturity of 12 months.The Brazilian real deviation is highlighted in dark blue. The light blue lines representdeviations of the control group: Australia, Canada, Chile, Czech Rep, Euro, Hungary, India,Israel, Japan, Mexico, Philippines, South Korea, and Turkey. The criteria used to selectthe control group are explained in Section 2.0.1.1. The program was launched on Aug 22,2013, represented by the traced vertical line. On the right-hand side, the area filled in greyrepresents the cumulative swap interventions, started on May 31, 2013, and lasting until theend of the program, on March 31, 2015.

unchanged, the on-shore dollar rate, known as Cupom Cambial, must increase.As we explain later, in Section 3, the CIP in Brazil can be expressed as theCupom Cambial less the LIBOR. Then, selling USD on the forward marketwidens the parity. Through this mechanism, we expect a similar effect on spotmarket interventions.

We believe our study is relevant in two ways. First, measuring FXIspillovers is a relevant contribution, since FXI is widely and frequently adoptedfor several countries, especially Brazil. The amount of resources applied in thispolicy is also relevant. Second, understanding the components behind the CIPdeviations is quite relevant since these deviations represent a market failure,which implies inefficiency and loss of welfare [Cerutti et al. (2019)]. In themodel developed by Amador et al. (2019), for instance, the CIP deviation isthe measure of welfare loss incurred by a central bank that pursues an exchangerate policy. Still, for the special case of Brazil, this deviation represents thecost incurred by the BCB of carrying forward international reserves.

Our main contribution to the literature is to capture the dynamics of

DBD



this impact. Garcia and Volpon (2014) was the first to document the contem-poraneous relationship between forwarding interventions and CIP deviations.Walker (2019) build up a static model showing the asymmetry between thespot and forward prices during the intervention in the forward market. Theauthor also shows contemporaneous evidence for Mexico and Peru. We con-tribute by bringing new methodologies for this literature, the ArCo and LocalProjection, and understanding the long-term impact of such interventions.

The rest of the paper is organized as follows. In the sequence, there isa literature review. In Section 2, we present the ArCo and Local Projectionmethodology. In Section 3 we present the data and how we construct the CIP.In Section 4 we show the results for both methodologies and in Section 5 weconclude.

1.0.1Related Literature

In the wake of the 2008 GFC, several emerging market economiesadopt FXI to shield their currencies against the capital flows caused by theQuantitative Easing (QE), reviving the question of whether are these sterilizedinterventions effective or not.

Several papers in the last decade have addressed this question. Chamonet al. (2017), which share the ArCo methodology and the major swap interven-tion program used in the present study, conclude that the swap interventionsled to an appreciation of the Brazilian real in 10 percent; Kohlscheen and An-drade (2014) use high-frequency data and also find a significant impact onUSDBRL, for the years of 2011 and 2013.

Dominguez (2019), in turn, generalizes the study by looking at severalemerging economies, in a larger period. The results are, however, inconclusive,and the effectiveness depends on the period and the country under analysis. Agreat contribution of Dominguez (2019) regards the methodology of definingFXI, as many central banks do not display these data explicitly. This subjectis also explored by Blanchard et al. (2015), using their results to categorize thecountries between floaters and de facto interveners.

Nedeljkovic and Saborowski (2019) address a slightly different questionregarding the relative effectiveness in intervening on the spot or the forwardmarket. Looking at the Brazil interventions, for the period 2008-2013, theauthors find similar effects on the USDBRL path using both instruments. Thereason for Nedeljkovic and Saborowski (2019) use Brazil data is because BCBused both instruments simultaneously in that period, although it not includethe swap program. Oliveira (2020) reaches the same result, the effectiveness

DBD



of both instruments, employing GMM estimations, using BCB interventionsfrom 2006 until 2016. These findings motivated us to investigate if spot andforward interventions have the same effect on CIP deviation.

A more recent literature study to what extent these FXI producespillovers on other economic variables. Hofmann et al. (2019) study the ef-fect of FXI on domestic credit in Colombia. Using sectoral microdata, theyfind that the purchase of dollars dampens the flow of new domestic corporateloans in the country. Gonzalez et al. (2019) shows that FXI accomplished byBCB diminished by half the effects of the GFC on local credit supply andemployment.

The papers we follow closer are Walker (2019) and Garcia and Volpon(2014). Investigating the interventions that occurred in 2008-2009 and 2013-2014, Garcia and Volpon (2014) conclude that domestic non-deliverable for-wards (DNDF) interventions provide BCB’s an alternative intervention strat-egy by providing a hedge for the market players. The authors find empiricalevidence of a contemporaneous relationship in Cupom Cambial and this kindof intervention. Walker (2019) creates a static model to analyze the differencesbetween both instruments and concludes that, although they have a similarimpact on the spot price, their effects are different on the central bank’s netforward position. According to the author, "A change in the central bank’s for-ward position has a greater effect on the forward rate than on the spot rate"which, in turn, widens the CIP deviations. It is in this literature where ourmajor contribution, since our paper is the first to capture the dynamic effectsof FXI on CIP deviations, up to our knowledge.

Finally, our paper is also related to the CIP deviations studies, thegrowing literature that tries to understand the phenomenon occurring in theadvanced economies after the GFC. The main question to be tackled in thesepapers is what are the determinants of these deviations? Avdjiev et al. (2019)highlights the role of the strong dollar on the banks’ balance sheet, diminishingtheir appetite for arbitrage possible deviations. Du et al. (2018) deservea prominent role in financial regulation and how they pressure the banks’balance sheet. In the years that followed the GFC, tighter regulations on thebanking system diminished the banks’ capacity to arbitrage the deviations. Theauthors also observed that countries with lower nominal interest rates showedhigher deviations from CIP. Cerutti et al. (2019) attribute the deviations notonly for regulatory changes, after the GFC, but for multiple macro-financialfactors. Beyond the prominent role exercised by a stronger dollar after theGFC, the authors also find statically significant regulatory and unconventionalmonetary policy effects after 2008, although these factors are temporary and

DBD



vary across countries. In the Amador et al. (2019) study, a prominent roleis deserved to the zero lower bound environment. The inverse relationshipbetween nominal interest rates and the deviations is observed concomitantto a positive relation between deviations and the size of foreign reserves, formost of the G10 currencies. This empirical finding is attributed to the factthat monetary authority can sustain a positive deviation by accumulating asufficiently large position in foreign assets, through interventions on the foreignexchange market. Our contribution to this literature regards on empiricallycapture the dynamics of an FXI shock on the CIP deviations.

DBD


2Methodology

In order to capture the foreign exchange intervention dynamics on thecovered interest parity, we employ two different methodologies. First, wepresent the ArCo methodology developed by Carvalho et al. (2018). ArCobelongs to the counterfactual class of analysis, showing two main differenceswith the canonical Synthetic Control developed by Abadie et al. (2010): it ismore appropriated to deal with time series and ArCo does not require that theinterest variable relies on a convex hull combination of peers.

Then, we present the assumptions we use to build the impulse responsefunctions using Local Projection of Jordà (2005). Still, we follow closely Romerand Romer (2010) and Ramey and Zubairy (2018) by using identified shockson the IRF construction.

2.0.1Artificial Counterfactual (ArCo)

To implement the methodology in R language, we use the packagedeveloped by Fonseca et al. (2017)1.

Let yit = (y1it, . . . , y

qiit )′ ∈ Rqi be the set of all observable variables, where

i indexes the countries, t = 1, . . . , T the days and qi the number of variablesper country. Let also Dt be a binary variable assuming one when the country isexposed to the intervention and zero otherwise. For the specific case of Brazil,we have yBR,t = Dty(1)

BR,t + (1−Dt)y(0)BR,t.

Then, the interventions can be represented2 by

y(1)BR,t =

y(0)BR,t, t = 1, . . . , T0 − 1

δt + y(0)BR,t, t = T0, . . . , T

(2-1)

where δtTt=T0 is a deterministic sequence and T0 represent the period of

intervention. As y(0)BR,t is not observable after T0, consider the approximating

model for yt.1For implementation of the ArCo package in the R language, we follow closely the Fonseca

et al. (2018) manual.2The methodology developed by Carvalho et al. (2018) allows for a more general

framework. The estimation can be done for some arbitrary function of y, which will dependon the moment of interest. We use the simplest specification since we are only interested ineffects on the mean.

DBD


Chapter 2. Methodology 18

y(0)BR,t =M(Y0,t, θ0) + εt, t = 1, . . . , T (2-2)

The Y0,t in Equation 2-2 correspond to the collection of all the untreatedunits’ observable Y0,t = (y2,t, . . . ,yn,t), where n represent the number ofcountries in the sample. Our main objective is to consistently estimate δt =yBR,t − M(Y0,t, θ0), for t = T0, . . . , T . Carvalho et al. (2018) demonstratethat θ0 is a consistent estimator for θ0 using only observations before theintervention.

ArCo methodology allows for quite general choice of M(Y0,t, θ0). Anatural choice is the usual ordinary least square (OLS), which we use toestimate the daily regressions. However, when we look at weekly data, inour robustness test, the pre-treatment sample becomes quite short. We tacklethis problem by estimating M(Y0,t, θ0) using LASSO, which enables us towork in a high dimension environment. We use the R package developed byVasconcelos (2017), which selects the λ in LASSO by looking at the lowestBayesian Information Criteria (BIC).

In this paper, the particular y(0)BR,t we are interested in is the CIP

deviations from the Brazilian real against the American dollar. However, aproblem we face in estimating this kind of equation is the difficulty in explainsthese deviations. Even controlling for several different variables related to theparity, the poor fit in the pre-intervention observations may affect the qualityof the post-intervention estimation.

We then ran a robustness exercise, in the spirit of Fama (1984), and alsoused in some recent papers [Cerutti et al. (2019), Engel et al. (2019)], consistingin split the CIP between the Forward Premium (FP) and the Interest RatesDifferential (IRD) to estimate equations like Equation 2-3 bellow3.

FP(0)BR,t = β IRDBR,t +M(Y0,t, θ0) + εt, t = 1, . . . , T (2-3)

Fama (1984) adopted this strategy to verify whether the CIP holds ornot for a given pair of countries, by regressing the FP on the IRD. If the parityholds, it is expected an estimated intercept equal to zero and the estimatedIRD coefficient β equals one4. The advantage of this approach, in our context,is the higher explanatory power compared to the CIP alone, as IRD explainsmost of the FP.

The null hypothesis of interest assumes the form of Equation 2-4.3By definition, CIP is equal to the interest rate differential minus the forward premmium.

See the Section 3 for further details.4The intercept is not showed explicitly in Equation 2-3 because the notation assumes

thatM(Y0,t, θ0) contains the intercept term.

DBD



H0 : ∆t = 1T − T0 + 1

T∑t=T0

δt = 0 (2-4)

Carvalho et al. (2018) build the Equation 2-4 statistics using δt. Moreover,the authors construct the confidence interval for ∆t, which we use to evaluateif the program delivered a higher CIP deviation after the intervention, onaverage, for a given confidence level.

Lastly, we are aware that using the Artificial Counterfactual methodol-ogy, a straightforward exercise would be to compare the results with the mosttraditional counterfactual methodology, the Synthetic Control (SC), developedby Abadie et al. (2010). However, the Brazilian CIP deviations violate a crucialhypothesis of SC by lying outside the convex hull formed by the peers. Figure1.1 illustrate this fact: the 12-month Brazilian CIP deviations is higher thanany peer deviation during several periods. This violation compromises the SCresults since no convex combination of CIP deviation from the control groupcan achieve the Brazilian deviation. Even if we could consistently estimateSC equations, Carvalho et al. (2018) points to the advantage in using ArCoinstead when working in a time series environment, since the Abadie et al.(2010) methodology do not account for the time dimensions of the data.

2.0.1.1Identification

A crucial assumption required by the ArCo is that peers cannot beaffected by the intervention5. In our context, the Brazilian Central Bank (BCB)handled these FX interventions to smooth movements in the USDBRL causedby the taper tantrum.

In this context, for a country belongs to the control group it is sufficientthat no FX intervention had been ruled by the respective central bank duringthe period in the analysis. An implicit hypothesis we also assume is thatinterventions in Brazil do not affect the parity in other countries in the controlgroup, which seems reasonable since Brazil does not represent a large financialmarket.

However, assessing whether or not a country has intervened in the foreignexchange market is not a trivial exercise. As discussed by Dominguez (2019),central banks are not always clear how and when they intervene in their localforeign exchange market.

Then, to construct a list of countries eligible for the control group, wemix countries used by Dominguez (2019) and Chamon et al. (2017). The onlycountry used by both papers and not here is Colombia. According to Hofmann

5In fact, this hypothesis is also required by the Synthetic Control of Abadie et al. (2010).

DBD



et al. (2019), the Colombian monetary authority ruled pre-announced FXauctions during 2011 and 2014, aiming to increase its international reserves.A caveat regarding Turkey is the fact that the Central Bank of Turkey hadintervened in a single day selling almost 3 bn USD on Jan 23, 2014. We donot believe this single auction would affect our identification because we seeas an isolated event since no intervention has occurred after this day andprevious interventions occurred only in Jan 2012. Moreover, we see Turkeyas an important peer of Brazil, mainly due to their similar interest ratesdifferential, so that we decide to keep the country in the analyses, followingDominguez (2019) and Chamon et al. (2017).

Although both papers deal only with emerging market economies, weadd to the control group some developed economies usually studied in theCIP deviations literature, also known as the G106 currencies, since our dailydata allows a higher number of variables. For the case of weekly data, theLASSO circumvent any trouble of adding more variables in the estimation.After analyzing the actions of its central banks during 2013 and 2015, weconclude that only Australia, Canada, Eurozone, and Japan are eligible.

Considering all the points mentioned above, the thirteen countries thatcompose the control group are: Australia, Canada, Chile, Czech Republic,Eurozone, Hungary, India, Israel, Japan, Korea, Mexico, Philippines, andTurkey.

Finally, the identification hypotheses also affect the pre-treatment period.During 2012, the BCB intervened through swaps for several days. In May2012, it sold around 5.5 bn USD; in June, more than 10 bn USD and betweenNovember and December, more 5.5 bn USD. Then, we choose to restrict ouranalysis for the year of 20137 as we discuss in detail in the next section.

2.0.2Local Projection

A disadvantage in estimating the ArCo statistics regards the lack ofinformation about the exact effect of an FXI on the CIP deviation. Forestimating the aggregate average impact, ArCo even uses the interventionobservations.

Aiming to tackle this problem, we use the Jordà (2005) Local Projection(LP) to estimate impulse response functions (IRF’s) of CIP deviations’ reaction

6According to Du et al. (2018), the G10 currencies correspond to Australian dollar,Canadian dollar, Swiss franc, Danish krone, British pound, Euro, Japanese yen, Norwegiankrone, New Zealand dollar and Swedish krona.

7There was a single intervention ruled on March 27, 2013, when BCB sold almost 1 bnUSD, which we consider as an isolated event. Nevertheless, we test our results excluding thisspecific day of the sample, and the results remain unchanged statistically.

DBD



to an intervention. We estimate the responses in R language using the packagedeveloped by Adämmer (2019).

Among the advantages of LP cited by Jordà (2005), we highlight therobustness to misspecification and the inference, which follows the standardOLS procedures. The only concern regarding the inference is the error serialcorrelation. Then we use the Newey and West (1987) robust standard errors,as recommended by Jordà (2005).

We follow closely the approach of Romer and Romer (2010) and Rameyand Zubairy (2018) by identifying the shock. Let consider the followingframework

CIPDt = α + βCFt + ΘCRt + εt (2-5)where CFt represents the capital flows and CR the country risk in the period t.CIPDt is the deviation observed from the parity in the same period. Obviously,the effects are not only simultaneous but for simplicity, we ignore the dynamicsfor now.

Because the CIP is an arbitrage condition, it is closely related to thecomplete market hypothesis. Any restriction on the capital flows may affectthe players’ capacity to negotiate the assets involved in the parity, distortingprices and leading to deviations. The second component is usually attributedto the CIP deviations observed in emerging markets currencies. It contemplatesrisks of credit, convertibility, liquidity, and sovereign defaults.

The term εt includes other factors that transiently affect the deviations.Recent literature assigns as candidate causes the banking regulations, theUSD strengthening and the zero lower bound environment for the deviationsoccurred after 2008 GFC, as discussed in Section 1.0.1, and might be correlatedwith each other.

εt =K∑

i=1εi

t (2-6)

At the same time, most of these factors also affect the capital flow amongcountries. Additionally, FXI is a measure adopted by central banks to correctdysfunctionalities on the exchange market caused by abrupt capital flowschange. In the taper tantrum context, as mentioned earlier, FXI was adoptedto circumvent the massive capital outflow caused by the quantitative easingmeasures’ withdraw, as one example. Then, consider the following specificationfor the capital flow variable.

CFt =K∑

i=1εi

t + FXIt (2-7)

where εit are the same as before. FXIt represents the foreign exchange inter-

DBD



vention, which in our case assumes interventions in the forward and the spotmarket. This specification implies that interventions are exogenous to the otherfactors that transiently affect the deviations. We believe this hypothesis holdstrue since the FXI aims only to give foreign exchange market liquidity. Thisnecessity is triggered by reasons as internal political events and/or marketstress observed along a day, in events when BCB decides to sell dollars. Thepurchase of foreign currencies is motivated by good market conditions and themonetary authority desire to accumulate reserves8.

Then, replacing 2-7 and 2-6 on 2-5

CIPDt = α + β

[K∑

i=1εi

t + FXIt

]+ ΘCRt +

K∑i=1

εit (2-8)

Adding the dynamics interaction and rearranging the terms, the speci-fication takes the form of Equation 2-9 below. We use as proxies for countryrisk log difference of 5-year CDS and VIX, and the bid-and-ask spread forUSDBRL. We control both CIPDt and CRt with 5 lags, capturing the lastweek’s movements. The IRF is then defined as the sequence of all coefficientsrelated to the shock variable βhH

h=0, where h index the horizons.

CIPDt+h = α + Ωh(L)CIPDt−1 + Θh(L)CRt−1 + βhFXIt+h + ut (2-9)

where ut is a functions of εit+h and it is uncorrelated to the interventions.

Although the differences in IRF construction, Plagborg-Møller and Wolf(2019) argues that LP and VAR estimations are the same for a less restrictiveset of hypotheses than the correct modeling of DGP, as proposed by Jordà(2005). According to the authors, the only trade-off researchers face betweenthe two techniques is that LP produces low bias estimations, while VAR hasa lower variance. Plagborg-Møller and Wolf (2019) still show the estimationsare equivalent up to a settled horizon.

Summarizing, we claim that the BCB do not look at the CIP when decideto intervene or not, it only consider movements on the exchange rates causedby liquidity problems, and the effect is given through this channel, of capitalflows movement. Note this is not the case when studying the FXI impacts ofthe exchange rates [Chamon et al. (2017), Hofmann et al. (2019)].

8For a detailed discussion about the hypothesis that must hold for consistently estimatethe dynamics, see Stock and Watson (2018).

DBD


3Data

The currencies and interest rates data used to calculate the CIP devia-tions come from Thomson Reuters. We also use data of 5-year CDS (CreditDefault Swap) and equity indices in ArCo. A complete list of the variables canbe found in Appendix A. In the robustness exercise, we use weekly data ofequity and bond flows per country, in USD, from Haver Analytics.

To estimate the Local Projection impulse response functions, we use dailydata of swap interventions, provided by the AC Pastore, and spot interventionsseries from BCB website, from 2009 until 2020.

3.0.1The Program of Daily FX Intervention

On May 22, 2013, after six years of Quantitative Easing (QE), the thenFED-chairman Ben Bernanke first revealed the FOMC’s plan to step down thepace of the ongoing expansionary monetary policy, in what becomes known asthe taper tantrum.

For the emerging markets, like Brazil, the end of the QE represented amassive outflow of capital. In the specific case of the Brazilian real, it wasfollowed by a 10% depreciation in the month after the announcement, thestrongest effect among the emerging markets economies, as pointed out byEichengreen and Gupta (2015). Strong currency depreciation could negativelyimpact the local private sector, which held a high debt in dollars, hired in thecontext of capital inflow followed by the GFC, as explained by Domanski et al.(2016). Then, to strike the consequences of the capital outflow, the BrazilianCentral Bank began to intervene in the FX market, using forward derivativeinstruments.

Among the advantages of intervening in the forward exchange market,Nedeljkovic and Saborowski (2019) cite, as the most relevant at that time, thefact that forward interventions do not imply in international reserves changes,because no payments are required at the time the contracts are negotiated. Atthe maturity, however, the Central Bank pays the USDBRL variation of theperiod plus the ex-ante on-shore dollar rate, the Cupom Cambial. On the otherhand, the counterpart pays the ex-post variation of domestic interest rate DI.

DBD


Chapter 3. Data 24

Thereby, the Central Bank offers a hedge against USDBRL variations to themarket players without the necessity of buying spot dollars.

On May 31, 2013, BCB started to sell swap contracts in a discretionaryway: the central bank made auctions only in days it thought was necessary.However, these interventions had not achieved the results the monetary au-thority was expecting, and on August 22 BCB announced a major programof daily interventions until, at least, the end of 2013. Every working day, theBCB sold USD 500 mn worth of currency forward through swaps. Moreover,the program also contemplates a weekly USD 1 billion repurchase agreement(repo) auction offer, on Fridays.

Due to its good performance, the program was extended three times inthe following months, ending only on March 31, 2015. These extensions arefrom our particular interest because it will guide the way we partition thesample. First, we separate the pre-intervention sample in two periods: the pe-riod where no intervention was ruled and the period where only discretionaryinterventions happened. The objective is to understand if there is an effect oninterventions or in BCB compromise in intervening. We interpret these twopossible effects as analogous to the portfolio balance and signaling channelsdescribed by Sarno and Taylor (2001). Then we analyze if the extensions an-nouncements had any impact on CIP deviations, splitting the post-interventionsample, according to the respective dates.

In Figure 3.1 we show the sub-samples we work with. We label eachone as "Sample". The group where no intervention was ruled and the periodpre-program announcement are represented by Samples 1 to 5 and Samples 6to 9, respectively, in Figure 3.1. Sample 1 we use to evaluate the efficiency ofintervening or not while in Sample 6, we evaluate the efficiency of the program.Note that the post-intervention period is the same for Sample’s 2 and 6, 3 and7, 4 and 8, and 5 and 9. We highlight the dates of interesting, replicating inthe top, middle and bottom of the table.

One implication of this subsampling is the number of observations in eachSample. Using daily data, Samples 1 to 5 have 100 days, while Sample 6 to 9has 159. When we look at the weekly data however, the number of observationsshrinks to 21, in the first group, and 33 and the second group. This fact leadsus to estimate weekly models using LASSO in the ArCo first step, as we willexplain in Section 4.

DBD


Chapter 3. Data 25

Figure 3.1: Periods of Analysis

The intervention day for Sample 1 to 5 is when BCB began to intervene in a discretionaryway, on 31 May 2013. The post-intervention period for Sample 2 is the program announce-ment; for Sample 3 is the first extension, on 18 Dec 2013; Sample 4 is the second extension,on 24 Jun 2014; and Sample 5 is the last extension, on 7 Jan 2015. For Sample 6 to 9,the intervention day is when BCB announces the major program of daily swap sales. Thepost-intervention periods are analogous to the Sample 1 to 5.

3.0.2CIP Deviations

According to the covered interest parity condition, for any two currencies,their interest rate differential for some maturity k must be equal to theirforward premium, evaluated as the difference between the forward exchangerate, of maturity k, and the spot exchange rate. As it consists of a no-arbitragecondition, any deviation represents a profitable risk-free opportunity, since allfuture values are pre-determined at the time of the transaction.

Let it,t+k represents the domestic interest rate of maturity k and i∗t,t+k

represents the foreign interest rate; St is the spot exchange rate and Ft,t+k thecorresponding forward price, both defined as the amount of domestic currencynecessary to buy one dollar. Then, the CIP condition can be represented as

St(1 + it,t+k)k = Ft,t+k(1 + i∗t,t+k)k(1 + xt,t+k)k (3-1)where xt,t+k represents the deviations from CIP for a maturity k. It is

usual in the literature - and more convenient - to represent the equation in logterms.

ln(1 + xt,t+k) = ln(1 + it,t+k)− ln(1 + i∗t,t+k) − ρt,t+k (3-2)The first component of Equation 3-2 represents the interest differential

and ft,t+k−st

k≡ ρt,t+k represents the forward premium. Note that we adjust the

maturity in the forward premmium.Regarding the variables used to calculate the CIP deviations, we follow

Du et al. (2018) and use the interbank interest rates, with i∗t,t+k representingthe USD LIBOR. Because the LIBOR exists for the maturity of up to one

DBD


Chapter 3. Data 26

year, we restrict our analysis to the maturities k = 1, 3, 6, 9, 12 months.1

Although the calculation of CIP for developed economies is straightfor-ward, when looking at emerging economies some problems arise. Chile doesnot have usual forward currencies contracts, so we use NDF (Non-DeliverableForward) contracts for calculating their deviations. Still, Brazil has issues re-garding currency forward derivatives and interest markets. As Brazil holds aprominent position in our analyses, we reserve the next subsection to explainhow its future exchange rate market works.

3.0.2.1Brazil

Although Brazil has a usual forward dollar market, the liquidity in thismarket is concentrated only on the shorter maturity of one month2, which ariseproblems when we look at longer maturities, as is usual in the CIP literature.

For historical reasons, Brazil has developed a market of on-shore dollarinterest rates, known as Cupom Cambial, and it can be defined as the yield, indollars, for an on-shore investment applied in Brazilian reais (BRL). Then, theprice of Cupom Cambial consists of the local interest rate - DI3 - minus theexpected depreciation of the local currency against de dollar. For this reason,Cupom Cambial is commonly used to extract the market expectation of futureBRLUSD quotations.

In our study, we follow previous studies in calculating the CIP deviationfor Brazil using only the Cupom Cambial and the LIBOR in dollars. Itis important to note that using the Cupom Cambial instead of the futuredollar does not affect the CIP deviation measure since market operators couldarbitrage any wedge between these two markets. Cieplinski et al. (2014) definesthe Cupom Cambial ict,t+k of maturity k as

(1 + ict,t+k)k = (1 + it,t+k)k

Ft,t+k

St

(3-3)

Then, from equations 3-2 and 3-3, we can define the CIP deviation inBrazil, against the US dollar, as

ln(1 + xt,t+k) = ln(1 + ict,t+k)− ln(1 + i∗t,t+k) (3-4)1The study of CIP for longer maturities is calculated by the spread on the cross-currency

basis swap. See Du et al. (2018) for details.2Actually, the one-month forward market is more liquid than the spot market in Brazil,

since only banks can operate in the spot market, and even the prices are formed first in theforward market than the spot market, as shown by Ventura and Garcia (2012).

3Depósito Interbancário, the domestic interbank deposits.

DBD


Chapter 3. Data 27

where i∗t,t+k is the LIBOR quoted in dollars. Note that, for extractingthe Brazilian forward premium of maturity k, we can just subtract the localinterest rate DI from the Cupom Cambial.

Table 3.1: CIP Deviations - Descriptive Analyses

Country 1 Month 3 Month 6 Month 12 MonthMean SD Max Min Mean SD Max Min Mean SD Max Min Mean SD Max Min

Australia -0.05 0.75 4.67 -4.39 -0.01 0.26 1.63 -1.55 0.02 0.15 0.88 -0.77 0.05 0.12 0.57 -0.43Brazil 0.88 0.57 3.57 -0.93 0.91 0.38 2.14 0.04 1.11 0.38 2.35 0.32 1.22 0.48 2.75 0.39Canada 0.18 0.66 3.72 -2.86 0.15 0.23 1.42 -0.77 0.14 0.12 0.84 -0.27 0.04 0.09 0.46 -0.96Chile -0.22 0.53 0.82 -2.34 0.08 0.34 0.96 -1.05 0.33 0.25 0.96 -0.30 0.24 0.30 0.81 -3.67Czech Rep 0.28 0.61 3.56 -5.14 0.29 0.21 1.42 -1.42 0.31 0.14 0.95 -0.48 0.29 0.15 0.85 0.03Euro -0.17 2.78 11.47 -27.86 0.02 0.91 4.04 -8.39 0.07 0.46 2.09 -3.98 0.09 0.25 1.12 -1.79Hungary 0.59 0.87 5.46 -3.40 0.65 0.39 2.32 -0.60 0.68 0.53 6.43 -0.28 0.45 0.22 0.96 -0.45India -0.31 0.92 5.08 -4.87 0.21 0.63 2.75 -1.75 0.56 0.69 2.86 -1.03 0.99 0.70 3.54 -0.27Israel 0.13 0.63 3.55 -3.01 0.18 0.26 1.22 -0.97 0.13 0.19 0.83 -0.54 0.00 0.17 0.65 -0.48Japan 0.28 0.59 3.67 -2.90 0.25 0.21 1.53 -0.80 0.26 0.13 0.94 -0.23 0.14 0.12 0.56 -0.17Mexico 0.10 0.87 3.00 -6.00 0.30 0.32 1.52 -1.87 0.28 0.19 1.18 -0.75 0.14 0.12 0.48 -0.49Philippine 1.48 0.90 4.83 -0.85 0.08 0.47 1.58 -1.13 -0.02 0.41 0.95 -1.01 0.01 0.43 1.01 -0.86South Korea 0.24 0.47 2.61 -6.48 0.38 0.26 1.16 -1.90 0.49 0.18 1.05 -0.61 0.60 0.15 1.23 0.09Turkey 0.19 1.82 5.17 -31.69 0.49 0.79 2.26 -9.02 0.53 0.64 2.19 -4.11 0.55 0.50 1.59 -1.67

CIP deviations are shown in % per year. We use daily data from the day BCB begin tointervene, on May 31, 2013, until the last intervention, on March 31, 2015. All data iscollected from Thomson Reuters Datastream. See Appendix A for further details.

In Table 3.1, some descriptive statistics of the CIP deviations are shownfor every maturity and country in the sample. Some patterns can be observedin these statistics.

First, the shorter maturities are more volatile than the longer ones.Second, longer maturities tend, in general, to have a higher mean. Thesetwo facts are compatible with previous papers that analyze the CIP deviationphenomenon in the G10 currencies, as in Du et al. (2018) and Cerutti et al.(2019). Finally, emerging economies have shown, in general, a higher deviation.This is in line with the idea that there exist risks in investing in these countries,mainly country risk and convertibility risk since these markets are less liquid ingeneral. For the case of Brazil, these facts are explained by Garcia and Didier(2003).

3.0.3Interventions

In the last ten years, the Brazilian Central Bank has used severalinstruments to intervene in the FX markets. After the 2008 GFC, it tookadvantage of the high inflow of capital to build up reserves, by purchasingAmerican dollars in the spot market. This movement lasts until May 2012 andwas responsible for an increase of 180 USD bn in the Brazilian internationalreserves4, as shown in light red in Figure 3.2.

4On 9 May 2011, the international reserves accounted for 191 USD bn and 2 May 2012,for 374 USD bn, according to Brazilian Central Bank.

DBD


Chapter 3. Data 28

Swaps interventions are shown in blue. The major program is representedthe dark blue, between 2013 and 2015. Other interventions via swap were alsoused along with our sample. At the end of 2019, the BCB took advantage ofits high international reserves level to operate by selling USD dollar in thespot market meanwhile it liquidate open swap contracts. More recently, in thecapital outflow caused by the Covid-19 pandemic, it uses swap and spot salesto provide liquidity to the market.

Figure 3.2: Foreign Exchange Interventions

Figure 3.2 shows the several instruments used by the Brazilian Central Bank to intervenein the FX market, as well as the direction of the interventions. The frequency is daily. Thesource is the BCB website.

Observing Figure 3.2, several specifications emerges for test with LocalProjection. Beyond the primary question of whether forward interventionsaffect or not the CIP deviations, we also test if the reverse swap has theopposite effect. Still, we can test the Walker (2019) hypothesis that spotintervention does not affect the parity, for both cases of purchasing or sellingforeign currency.

In this spirit, we delimit periods for testing the efficiency of eachinstrument used by the BCB, to match the periods these instruments were

DBD


Chapter 3. Data 29

used in fact. For testing the swap intervention effect, we stick to the major2013 program, explained earlier, and the auctions made after the Covid-19crisis onset. Spot purchases are restricted to the years of 2009 to 2012. Spotsales were used only recently. Regarding the swap reverse, we have two distinctdata, as can be visualized in Figure 3.2: a period in 2016, which we call SwapReverse - 1, and the second period, used together with the spot sales morerecently, Swap Reverse - 2.

All these samples are detailed in Table 3.2. As can be seen in theObservations column, the sample size varies among each instrument, whichleads to implications in the number of the horizon we estimate the IRF,considering the losses in terms of the degree of freedoms.

Table 3.2: Samples - Local Projection (LP)

Operation Begins Ends Observations HorizonsSpot Purchases 12 May, 2009 2 May, 2012 737 20Spot Sales 23 Aug, 2019 23 Dec, 2019 84 10Reverse Swap - 1 1 Jul, 2016 8 Nov, 2016 89 10Reverse Swap - 2 21 Aug, 2019 20 Dec, 2019 85 10Swap 31 May, 2013 31 Mar, 2015 462 20Swap 2 13 Feb, 2020 15 May, 2020 61 10Repo Purchases 2 Jan, 2013 8 May, 2020 1806 20Repo Sales 2 Jan, 2013 8 May, 2020 1806 20

DBD


4Results

As mentioned in Section 2, we use two complementary approaches tocapture the dynamics of CIP deviations through the counterfactual analysis.First, using the CIP and the second, in the spirit of Fama (1984), regressingthe forward premium on the interest rate differential. We aim to better explainthe pre-intervention sample by using these approaches.

Explaining the CIP deviations in the first step of ArCo is not a trivialexercise. As discussed earlier, several papers are trying to explain which factorsare behind these deviations in the last few years. In this study, in a dailyfrequency data, this issue is accentuated. For this reason, we run tests forseveral different specifications, varying the exogenous controls, USD Index andVIX, and the variables common to all peers. For daily data, we test 32 differentspecifications, and for weekly data, 56, for each maturity and both CIP andForward Premium models. Combining different specifications affect slightly theresults, and we see this fact as robustness for our findings. The relation of themodels used can be found in Table B.1 and B.2, for daily and weekly datarespectively.

In Figure B.5 in the Appendix we show the Adjusted R-Squared foreach specification, in the pre-treatment period, when the parameters were es-timated. Two features from Figure B.5: first, longer maturities are better fitted,although the R2 are barely higher than 0.50; and two, the Forward Premiumequations have substantial higher coefficients of determination, reaching 0.8 inthe one-year maturity, showing the importance of our robustness test.

A drawback in working with daily data is the lack of informationabout capital flows, an important feature regarding CIP since deviationsfrom parity are expect to vanish through market players operating thisarbitrage opportunity. To circumvent this issue, we use weekly data of equitiesand bond flows of Haver Analytics. However, some pre-treatment samplesbecame significantly short, which lead us to estimate the pre-sample treatmentparameters by LASSO. The results are corroborated by this robustness testand are shown in the following.

In the last subsection, we show the IRF using Jordà (2005) LocalProjection. The results go in the same direction from ArCo models. Swaps

DBD


Chapter 4. Results 31

interventions widen the parity, and this result is symmetric depending on thepositions the BCB takes in the operation. Long-term maturities are moreaffected once again. The results regarding spot interventions, on the otherhand, points to no effect on the parity.

4.0.1Daily ArCo

The focus of our exercise is to look at the ArCo statistics. As explainedearlier, it accounts for the difference between the counterfactual and theobserved series on the post-treatment period. Carvalho et al. (2018) developedthe inference properties of ArCo statistics, which allow us to build up a 95%confidence interval and check if it is statistically different from zero.

Figure 4.2 shows the ArCo statistics, as well as its 95% confidenceinterval, for Sample 1 - comparing when BCB intervened or not - and Sample6 - evaluating the program announcement. Each color represents one of the32 specifications. Our interest is not to identify each specification but to shedlight on how the ArCo statistic changes depending on the variables we use tocontrol. The order, from left to right, follows the specification 1 to 32 in TableB.1.

The first thing to note is that the statistics are quite homogeneous alongwe change the specification. Looking at the CIP in Sample 1, we see that theinterventions ruled between May 31, 2013, until the end of the program doesnot seem to affect the CIP deviations, excepting a slightly negative impacton the maturity of 3 months. Although there is an increase in the FP of 12-months, the IRD in the same period also increases monotonically, offsettingthis effect.

On the other hand, results from Sample 6 tell a quite different story.Despite some net positive deviation in the 1-month maturity, a strong andstatistically significant impact occurs in the 9 and 12-months maturities. Thedecrease in the forward premium goes in the same direction of the IRD increase,boosting the CIP deviation. The results for Sample 2, 3, 4, and 5 are quitesimilar do Sample 1. The same is true for Sample 7, 8, and 9 concerningSample 6. The CIP results for those Sample’s can be found in Figure B.1 andB.2, respectively, and are left in Appendix B. In addition to the OLS usedfor estimating pre-sample parameters, we also test LASSO in this daily datamodels and results remain the same, except by the heterogeneity among thespecifications which is attenuated. Our main goal in testing LASSO was theconcern about short pre-treatment samples1.

1In fact, the lowest sample we work with, discounted the degree of freedom, is 44, in

DBD



We interpret this stronger effect on longer maturity as a pattern relatedto the maturity of contracts offered by the Brazilian Central Bank. In Figure4.1, we plot the average maturity of the contracts offered on the same day,since BCB usually offers more than one type of contract 2. Along with theprogram, BCB offered longer contracts but never surpassing one year. Issuinglong-term contracts puts upward pressure on the maturity of 9 and 12 months,specially.

The pattern of offering longer maturities is discussed by Craveiro (2019).According to the author, BCB began to issue longer maturities contracts toavoid overload the Cupom Cambial market in the short term. The drawback inoffering long term contracts is an increase in the BCB exposition to variationsin the exchange rate, which would imply at a higher cost.

Finally, we also test the models excluding the day of March 27, 2013.As discussed in Section 2.0.1.1, in this day BCB handled a single intervention,which we claim was not a response to the taper tantrum shock, and thenshould not affect our identification strategy. The results change only in thesecond decimal place, which remains our findings unchanged.

Figure 4.1: Average Maturity of the New Offered Swap Contracts

Figure 4.1 shows the average of maturities, measured in days, from new offered swapcontracts by the Brazilian Central Bank. We use the average since for several days, BCBoffered more than one type of contract, with different maturities. The valleys that occurafter 2013, represent contracts offered with the same termination date. The cumulativeswap intervention is the same shown in Figure 1.1 replicated here for convenience.

Sample 1 to 5, when we estimate 56 parameters with 100 observations.2We plot the simple average, but we also calculated the weighted average, considering

the size of the contract. The patter is the same, but the series is noisier.

DBD


Chapter4.Results

33

Figure 4.2: Results - Daily ArCo

The colors represent an ArCo specification, as exposed in Table B.1. Each graphic represent a specific Sample division, as shown in Figure 3.1. The points representthe ArCo statistic, and the bars represent the 95% confidence interval. Looking at the Forward Premium models, a negative statistic imply that the observed FPis lower then the estimated counterfactual, which in turn implies in a lower CIP deviation in the absence of intervention.

DBD



4.0.2Weekly ArCo

The main drawback of working with daily data is the lack of informationabout exchange flows. We tackle this problem by using weekly data of flowsof dollars in bond and equity markets from Haver Analytics. To adjust theother variables for weekly frequency, we take the mean value of each one inthat specific week.

A question that might arise is why not to use the weekly accumulated CIPdeviations and the other variables instead of the mean. However, as explainedin Section 3, we use the first difference in the log for CDS, Equities Indices,VIX, and USD Index, and the difference between the log of bid and the logof ask, to evaluate the spread. Meanwhile, we use the level of CIP deviations,since theory assumes it is a stationary variable. Then, to accumulate thesevariables would bring us some inconsistency, which we circumvent by usingthe mean.

Figure 4.3 shows the results using weekly data. The results corroboratethe previous finding, by replicating the same pattern of longer maturities heldthe stronger effects. However, the analysis is less conclusive for the shortermaturities. The largest difference between daily and weekly exercises regardsthe homogeneity of the results: ArCo statistics are more sparse in the later, asbecame clear comparing Figures 4.2 and 4.3.

DBD


Chapter4.Results

35

Figure 4.3: Results - Weekly ArCo


DBD



4.0.3Local Projections

The main result of LP estimation is the fact that forward interventionsseem to affect the CIP deviations. Spot interventions, on the other hand, seemto have a limited impact on the deviations. Repo sales seem to impact thedeviations negatively, and the effect is more significant on the shorter maturity.Furthermore, the analysis among the maturities endorses the ArCo results: the12-month maturity is the most affected by swap interventions.

Figure 4.4: Swap Interventions Effect on CIP Deviations

The regressions follow Equation 2-9. The CIP deviations are controlled by the log differenceof 5-year Brazilian CDS, VIX and for the spread of bid-and-ask USDBRL in the spot market.We use 5 days lag in both endogenous and exogenous variables. The light blue line representsthe 95% CI, calculated using Newey and West (1987).

Figure 4.4 still points to a negative impact of reverse swap on CIPdeviations, indicating somewhat symmetry between the long and short BCBposition in this kind of instrument. The IRF’s plotted in Figure 4.4 contemplatethose reverse swap operations ruled occurred in 2016.

The main drawback faced when examining the intervention observationsis the fact the there is no much variation on the major swap program.However, we believe this fact plays in our favor since it implies larger estimatedconfidence intervals.

Since the onset of the Covid-19 and the social measures adopted for mostcountries to fight against the virus, there is an ongoing effort by economists tounderstand and measure the pandemic effects in the economy. Following this

DBD



effort, we add to our sample the recent interventions ruled by the BCB, until9 April 2020.

The main issue in analyzing this recent data is the small number ofobservations. The BCB first intervened on 13 Feb. Until 9 Apr, only 11interventions via swap has occurred, despite the relevant amount of USD 10.5bn. Then we had to shrink the horizon for 5 days and the lag for one day.We expect to follow the next interventions to evaluate the ongoing impactmore precisely. The details of intervention data can be found in Table 3.2. Theresults we have so far are shown in Figure 4.5.

Figure 4.5: Swap Interventions Effect on CIP Deviations - Covid-19


For both purchase and sales, CIP response to a spot intervention is moreirregular and not statistically different from zero, for most horizons. This resultmaintains for every maturity of parity we have studied, as shown in Figure4.6. Although this result guard similarities with Walker (2019), who suggeststhat any pressure put on the spot quotations are transmitted to the forwardprices equivalently, we observe this result cautiously. No major program of pre-announced spot interventions was indeed ruled in our sample, but differentlyfrom repo and swap interventions, the BCB action regarding purchase andsales of spot currency is clustered, as can be seen in Figure 3.2.

Despite any announcement, the purchases of spot currency from 2009 and2012 were ruled almost daily. At that time, the BCB used the massive inflow ofcapital, because of Taper Tantrum, to increase Brazilian international reserve.The same is true for spot sales interventions in 2H19. At that time, the BCBaimed to settled some open swap contracts by switching for spot currency, asexplained next. In both periods, despite any announcement, the pattern mightbe observed by the players, in the sense that these interventions have littleor no surprise effect and the impact on the CIP deviations was accumulated

DBD



throughout the whole period. The IRF’s for spot interventions can be seen inFigure 4.6 below.

Figure 4.6: Spot Interventions Effect on CIP Deviations


As mentioned, we deserve a special discussion for the concomitantinterventions of Reverse Swap and Spot Sales ruled in the second half of2019. The Central Bank liquidated open swap contracts through reverse swapoperations, by simultaneously selling spot currency. This operation aimed tore-balance the swap contracts portfolio and the liquidity in the spot market.Then it not represented an FXI as usual. Figure B.9 illustrate the IRF’s forthese interventions. The dynamic is completely different from those reverseswap operations ruled in 2016. The concomitant spot sales seem to offset anydeviation if not increasing, which seems to happen in the 1-month deviation.The spot purchase IRF’s are those shown in Figure 4.6. Then, there are reasonsto believe that the lack of effect on the CIP deviations could be driven by thisspecial intervention pattern.

Differently from the previous years, the Central Bank has used bothswap and spot sales in the Covid-19 crisis. However, a trade-off emerges inthe spot analysis, since the number of observation is even smaller than theswap interventions - only 6 days, summing USD 24.5 bn - and merging withthe 2019 data, the number of missing values strongly increases. We expect tocollect more data in the next months to correctly evaluate this impact.

DBD



Figure 4.7: Reverse Swap Effect on CIP Deviations - 2H2019


Figure 4.8: Repo Interventions Effect on CIP Deviations


As mentioned earlier, the repo lines interventions had a special placeon the 2013 major swap program. This line of credit is offered when theBCB observes some lack of liquidity in the spot market. In the 2013 programcontext, it was designed to correct some eventual liquidity dysfunctionality onthe spot market caused by the forward interventions. Then, one interpretationof the inverse pattern we observe in Figure 4.8 might be agents adjusting their

DBD



positions using the new credit lines, which closes the gap between forwardpremium and the interest rate differential.

Finally, the same IRF’s shown here for the maturities of 3, 6, and 9months were left in Appendix B.

DBD


5Conclusion

Covered interest parity (CIP) has been violated in emerging markets fora long time. For this reason, the absence of deviations to CIP was used as ameasure of perfect capital mobility. The 2008 GFC, and the tougher financialregulations that it entailed, generated deviations even in derivatives markets ofadvanced economies currencies, as documented by Du et al. (2018) and Avdjievet al. (2019) among others. Here, we contribute to this literature by analyzingthe effects of FX interventions on the CIP deviations.

Using the major intervention program ruled by the BCB, we find thatdaily forward interventions have increased the deviations in 35 bps in the12-month maturity, on average. The deviation increases along the maturitymonetary authority are issuing the contracts.

Local Projection shows a small contemporaneous effect which, however,remains along the time. This effect is also higher for longer maturities, the samepattern demonstrated in ArCo. Moreover, reverse swaps produce the oppositeeffect, evidencing symmetry between these operations.

Still, local projection indicated that spot interventions have a limitedimpact on the deviations. We observe this result cautiously since both spotpurchase and sale in our sample were ruled in specific conditions, which mightaffect the measure of this effect. Repo lines, on the other hand, seem to havethe opposite impact of swap interventions. One possible explanation for thisresult is the fact that this repurchase agreement corrects some liquidity issueson the spot market, adjusting the prices toward the parity.

There is some difficulty involved in linking the Local Projection’s impulseresponse functions results with ArCo due to the methodologies’ differentapproaches, despite the fact they point to the same directions. However, wethink the IRF sheds light on the dynamic of the interventions by indicatingsmall contemporaneous results which, however, remains in the middle term.

We believe our work contributes to FXI and CIP deviations literatureand left open some channels for further research on the interaction betweenthem.

DBD


Bibliography

Abadie, A., Diamond, A., and Hainmueller, J. (2010). Synthetic control methodsfor comparative case studies: Estimating the effect of california’s tobacco controlprogram. Journal of the American statistical Association, 105(490):493–505.

Adämmer, P. (2019). lpirfs: An r package to estimate impulse response functions bylocal projections. lpirfs: An R Package to Estimate Impulse Response Functionsby Local Projections. The R Journal. Forthcoming.

Amador, M., Bianchi, J., Bocola, L., and Perri, F. (2019). Exchange rate policiesat the zero lower bound. Review of Economic Studies. Forthcoming.

Avdjiev, S., Du, W., Koch, C., and Shin, H. S. (2019). The dollar, bank leverage,and deviations from covered interest parity. American Economic Review:Insights, 1(2):193–208.

Blanchard, O., Adler, G., and Carvalho Filho, I. (2015). Can foreign exchangeintervention stem exchange rate pressures from global capital flow shocks?Technical report, National Bureau of Economic Research.

Carvalho, C., Masini, R., and Medeiros, M. C. (2018). Arco: An artificialcounterfactual approach for high-dimensional panel time-series data. Journalof Econometrics, 207(2):352–380.

Cerutti, E. M., Obstfeld, M., and Zhou, H. (2019). Covered interest parity devia-tions: Macrofinancial determinants. National Bureau of Economic Research.

Chamon, M., Garcia, M., and Souza, L. (2017). Fx interventions in brazil: Asynthetic control approach. Journal of International Economics, 108:157–168.

Cieplinski, A., Braga, J., and Summa, R. (2014). Avaliação empírica do teoremada paridade coberta para a economia brasileira. Texto para Discussão, 22.

Craveiro, V. A. M. (2019). Efeito das intervenções do banco central do brasil nacurva de cupom cambial. MSc Dissertation in Macroeconomics and Finance -Department of Economics, PUC-Rio.

Domanski, D., Kohlscheen, E., and Moreno, R. (2016). Foreign exchange marketintervention in emerging market economies: what has changed? BIS QuarterlyReview September.

DBD


Bibliography 43

Dominguez, K. M. E. (2019). Emerging market exchange rate policies: Stabilizingor manipulation? IMF-CBC-IMFER Conference, Santiago, Chile.

Du, W., Tepper, A., and Verdelhan, A. (2018). Deviations from covered interestrate parity. The Journal of Finance, 73(3):915–957.

Eichengreen, B. and Gupta, P. (2015). Tapering talk: The impact of expectationsof reduced federal reserve security purchases on emerging markets. EmergingMarkets Review, 25:1–15.

Engel, C., Lee, D., Liu, C., Liu, C., and Wu, S. P. Y. (2019). The uncoveredinterest parity puzzle, exchange rate forecasting, and taylor rules. Journal ofInternational Money and Finance, 95:317–331.

Fama, E. F. (1984). Forward and spot exchange rates. Journal of monetaryeconomics, 14(3):319–338.

Fonseca, Y. R., Masini, R. P., Medeiros, M. C., and Vasconcelos, G. F. (2017).ArCo: Artificial Counterfactual Package. R package version 0.3-1.

Fonseca, Y. R., Masini, R. P., Medeiros, M. C., and Vasconcelos, G. F. (2018).Arco: A r package to estimate artificial counterfactuals. The R Journal,10(1):91–108.

Garcia, M. and Volpon, T. (2014). Dndfs: a more efficient way to intervene in fxmarkets? Texto para discussão No. 621.

Garcia, M. G. and Didier, T. (2003). Taxa de juros, risco cambial e risco brasil.Instituto de Pesquisa Econômica Aplicada (Ipea) - Pesquisa e PlânejamentoEconômico, 33(2).

Gonzalez, R., Khametshin, D., Peydró, J.-L., and Polo, A. (2019). Hedger of lastresort: evidence from brazilian fx interventions, local credit and global financialcycles. Economic Working Paper Series Working Paper No. 1648.

Hofmann, B., Shin, H. S., and Villamizar-Villegas, M. (2019). Fx intervention anddomestic credit: evidence from high-frequency micro data. BIS working paper.

Jordà, Ò. (2005). Estimation and inference of impulse responses by local projec-tions. American economic review, 95(1):161–182.

Kohlscheen, E. and Andrade, S. C. (2014). Official fx interventions throughderivatives. Journal of International Money and Finance, 47:202–216.

Menkhoff, L. (2013). Foreign exchange intervention in emerging markets: A surveyof empirical studies. The World Economy, 36(9):1187–1208.

DBD


Bibliography 44

Nedeljkovic, M. and Saborowski, C. (2019). The relative effectiveness of spotand derivatives-based intervention. Journal of Money, Credit and Banking,51(6):1455–1490.

Newey, W. K. and West, K. D. (1987). Hypothesis testing with efficient methodof moments estimation. International Economic Review, pages 777–787.

Oliveira, F. N. (2020). Empirical analyses of the effectiveness of interventions inthe foreign exchange market in brazil. Revista Brasileira de Finanças, 18(2).

Plagborg-Møller, M. and Wolf, C. K. (2019). Local projections and vars estimatethe same impulse responses. Unpublished paper: Department of Economics,Princeton University, page 1.

Ramey, V. A. and Zubairy, S. (2018). Government spending multipliers in goodtimes and in bad: evidence from us historical data. Journal of Political Economy,126(2):850–901.

Romer, C. D. and Romer, D. H. (2010). The macroeconomic effects of tax changes:estimates based on a new measure of fiscal shocks. American Economic Review,100(3):763–801.

Sarno, L. and Taylor, M. P. (2001). Official intervention in the foreign exchangemarket: is it effective and, if so, how does it work? Journal of EconomicLiterature, 39(3):839–868.

Stock, J. H. and Watson, M. W. (2018). Identification and estimation of dynamiccausal effects in macroeconomics using external instruments. The EconomicJournal, 128(610):917–948.

Vasconcelos, G. F. (2017). HDeconometrics: Implementation of several economet-ric models in high-dimension. R package version 1.0.

Ventura, A. and Garcia, M. (2012). Mercados futuro e à vista de câmbio no brasil:o rabo abana o cachorro. Revista Brasileira de Economia, 66(1):21–48.

Walker, C. (2019). Forward intervention. Foreign Exchange Intervention in InflationTargeters in Latin America.

DBD


AData Appendix

The following data are collected from Thomson Reuters Datastream plat-form.Exchange Rates. We use the Thomson Reuters as the source for the Aus-tralian dollar, Brazilian real, Canadian dollar, Czech koruna, Euro, Hungarianforint, Indian rupee, Israeli shekel, Japanese yen, Mexican peso, Philippinepeso, South Korean won, and New Turkish lira. For the Chilean peso, we useWM/Reuters as the source, the same for their NDF.

Interest Rates. All data are collected from the Thomson Reuters Datastreamplatform. The tickers of the respective variables are in parenthesis.

We use the USD LIBOR as the foreign interest rate when calculatingthe CIP. For the domestic interest rates, we use the Australian dollar deposit,Brazil DI-Pré fixed float, Canadian interbank offered rate (CIDOR), Chileaninterbank, Czech Republic interbank, Euribor, Hungary interbank, India de-posit rate, Tel Aviv interbank, Japan interbank (TIBOR), Mexico CETES,Philippine interbank call loan rate and Philippine treasury bill, South Koreainterbank and Turkey interbank.

When a particular maturity is not directly observed, we estimated vialinear interpolation. This is the case for the Japan interbank (TIBOR) for 3months, and for the Canadian interbank (CIDOR), Chilean interbank, TIBOR,Mexicos CETES and Philippine treasury bill for 9 months.

Equity Indices. S&P/ASX 200 (Australia), Bovespa (Brazil), S&P/TSXComposite Index (Canada), S&P/CLX IGPA CLP Index (Chile), Prague SEPX (Czech Republic), S&P Europe 350 (Euro), Budapest-BUX (Hungary),S&P BSE (SENSEX) 30 Sensitive (India), TA 125 (Israel), Nikkei 225 StockAverage (Japan), IPC (Mexico), SE I(PSEi) (Philippine), SE Composite-KOSPI (Korea) and BIST National 100 (Turkey).

Credit Default Swap. For most countries, we use sovereign 5-year CDS.We choose this maturity due to its liquidity. The exceptions are for Eurozone,which we use the 5-year Germany CDS, and for India, where the 5 year CDSseries starts only in March/2015 and we use the 5-year CDS of the Bank ofIndia, a state-owned bank, as a proxy.

DBD


BResults Appendix

Table B.1: List of the Models - Daily DataCovered Interest Parity Forward Premium

ArCo 1

No Exogenous Control

CIP ArCo 1

IRD

FPArCo 2 CIP + Equities Indices ArCo 2 FP + Equities IndicesArCo 3 CIP + CDS ArCo 3 FP + CDSArCo 4 CIP + Bid and Ask Spread ArCo 4 FP + Bid and Ask SpreadArCo 5 CIP + Equities Indices + CDS ArCo 5 FP + Equities Indices + CDSArCo 6 CIP + Equities Indices + CDS + Bid and Ask Spread ArCo 6 FP + Equities Indices + CDS + Bid and Ask SpreadArCo 7 CIP + Equities Indices + Bid and Ask Spread ArCo 7 FP + Equities Indices + Bid and Ask SpreadArCo 8 CIP + CDS + Bid and Ask Spread ArCo 8 FP + CDS + Bid and Ask SpreadArCo 9

USD Index

CIP ArCo 9

IRD + USD Index


VIX

CIP ArCo 17

IRD + VIX


USD Index + VIX

CIP ArCo 25

IRD + USD Index + VIX

FPArCo 26 CIP + Equities Indices ArCo 26 FP + Equities IndicesArCo 27 CIP + CDS ArCo 27 FP + CDSArCo 28 CIP + Bid and Ask Spread ArCo 28 FP + Bid and Ask SpreadArCo 29 CIP + Equities Indices + CDS ArCo 29 FP + Equities Indices + CDSArCo 30 CIP + Equities Indices + CDS + Bid and Ask Spread ArCo 30 FP + Equities Indices + CDS + Bid and Ask SpreadArCo 31 CIP + Equities Indices + Bid and Ask Spread ArCo 31 FP + Equities Indices + Bid and Ask SpreadArCo 32 CIP + CDS + Bid and Ask Spread ArCo 32 FP + CDS + Bid and Ask Spread

List of abbreviations: Covered Interest Parity (CIP), Forward Premium (FP), 5-year CreditDefault Swap (CDS). "Equities Indices" represent the stock market index of each country."Bid and Ask Spread" represent the log difference between the bid and ask spot dollar,for each currency. USD Index is the Trade Weighted U.S. Dollar Index, elaborated by theBoard of Governors of the Federal Reserve System (US). VIX is the CBOE Volatility Index,elaborated by the Chicago Board Options Exchange. Excepting the CIP, FP and IRD, allother variables are estimated by the log differential.

DBD


Appendix B. Results Appendix 47

Table B.2: List of the Models - Weekly DataCovered Interest Parity Forward Premium

ArCo 1

No Exogenous Control

CIP ArCo 1

IRD

FPArCo 2 CIP + Equities Indices ArCo 2 FP + Equities IndicesArCo 3 CIP + CDS ArCo 3 FP + CDSArCo 4 CIP + Bid and Ask Spread ArCo 4 FP + Bid and Ask SpreadArCo 5 CIP + Equities Indices + CDS ArCo 5 FP + Equities Indices + CDSArCo 6 CIP + Equities Indices + CDS + Bid and Ask Spread ArCo 6 FP + Equities Indices + CDS + Bid and Ask SpreadArCo 7 CIP + Equities Indices + Bid and Ask Spread ArCo 7 FP + Equities Indices + Bid and Ask SpreadArCo 8 CIP + CDS + Bid and Ask Spread ArCo 8 FP + CDS + Bid and Ask SpreadArCo 9 CIP + Equity Flow ArCo 9 FP + Equity FlowArCo 10 CIP + Bond Flow ArCo 10 FP + Bond FlowArCo 11 CIP + Equity Flow + Bond Flow + Equities Indices ArCo 11 FP + Equity Flow + Bond Flow + Equities IndicesArCo 12 CIP + Equity Flow + Bond Flow + CDS ArCo 12 FP + Equity Flow + Bond Flow + CDSArCo 13 CIP + Equity Flow + Bond Flow + Bid and Ask Spread ArCo 13 FP + Equity Flow + Bond Flow + Bid and Ask SpreadArCo 14 CIP + Eq. Flow + Bond Flow + CDS + EI + BA Spread ArCo 14 FP + Eq. Flow + Bond Flow + CDS + EI + BA SpreadArCo 15

USD Index

CIP ArCo 15

IRD + USD Index


VIX

CIP ArCo 29

IRD + VIX


USD Index + VIX

CIP ArCo 43

IRD + USD Index + VIX

FPArCo 44 CIP + Equities Indices ArCo 44 FP + Equities IndicesArCo 45 CIP + CDS ArCo 45 FP + CDSArCo 46 CIP + Bid and Ask Spread ArCo 46 FP + Bid and Ask SpreadArCo 47 CIP + Equities Indices + CDS ArCo 47 FP + Equities Indices + CDSArCo 48 CIP + Equities Indices + CDS + Bid and Ask Spread ArCo 48 FP + Equities Indices + CDS + Bid and Ask SpreadArCo 49 CIP + Equities Indices + Bid and Ask Spread ArCo 49 FP + Equities Indices + Bid and Ask SpreadArCo 50 CIP + CDS + Bid and Ask Spread ArCo 50 FP + CDS + Bid and Ask SpreadArCo 51 CIP + Equity Flow ArCo 51 FP + Equity FlowArCo 52 CIP + Bond Flow ArCo 52 FP + Bond FlowArCo 53 CIP + Equity Flow + Bond Flow + Equities Indices ArCo 53 FP + Equity Flow + Bond Flow + Equities IndicesArCo 54 CIP + Equity Flow + Bond Flow + CDS ArCo 54 FP + Equity Flow + Bond Flow + CDSArCo 55 CIP + Equity Flow + Bond Flow + Bid and Ask Spread ArCo 55 FP + Equity Flow + Bond Flow + Bid and Ask SpreadArCo 56 CIP + Eq. Flow + Bond Flow + CDS + EI + BA Spread ArCo 56 FP + Eq. Flow + Bond Flow + CDS + EI + BA Spread

List of abbreviations: Covered Interest Parity (CIP), Forward Premium (FP), 5-year CreditDefault Swap (CDS). "Equities Indices" represent the stock market index of each country."Bid and Ask Spread" represent the log difference between the bid and ask spot dollar, foreach currency. Equity and Bond Flow represent the Haver Analytics capital flows, in millionUSD, received by country. USD Index is the Trade Weighted U.S. Dollar Index, elaboratedby the Board of Governors of the Federal Reserve System (US). VIX is the CBOE VolatilityIndex, elaborated by the Chicago Board Options Exchange. Excepting the CIP, FP, IRD,Equity and Bond Flows, all other variables are estimated by the log differential.

DBD


AppendixB.

ResultsAppendix

48

Figure B.1: Results - Daily ArCo - Pre 1


DBD


AppendixB.

ResultsAppendix

49

Figure B.2: Results - Daily ArCo - Pre 2


DBD


AppendixB.

ResultsAppendix

50

Figure B.3: Results - Weekly ArCo - Pre 1


DBD


AppendixB.

ResultsAppendix

51

Figure B.4: Results - Weekly ArCo - Pre 2


DBD



Figure B.5: Results - Adjusted R-Squared

The colors represent the Samples correspond to the divisions, as shown in Figure 3.1. Sample1 to 5 correspond to the period where BCB intervened in a discretionary way, and Sample6 to 9, the period after the program was announced. Each point represents the adjusted R-Squared of specific specification, for the pre-treatment period, where the ArCo parameterswere estimated. CIP and FP correspond to the specifications were the CIP were estimateddirectly or split between forward premium and interest rate differential

DBD



Figure B.6: Swap Interventions Effect on CIP Deviations


Figure B.7: Swap Interventions Effect on CIP Deviations - Covid-19


DBD



Figure B.8: Spot Interventions Effect on CIP Deviations


Figure B.9: Reverse Swap Effect on CIP Deviations - 2H2019


DBD



Figure B.10: Repo Interventions Effect on CIP Deviations


DBD


DanielMalvezziDoine ForeignExchangeInterventionsandCovered ...€¦ · DanielMalvezziDoine...

Documents

Transcript of DanielMalvezziDoine ForeignExchangeInterventionsandCovered ...€¦ · DanielMalvezziDoine...